Local RAG (github.com)

A new workspace-based Retrieval-Augmented Generation (RAG) system has been announced, enabling users to chat about their private documents and codebases using local AI models. This experimental solution, built with React, FastAPI, Milvus, and Ollama, allows users to organize multiple projects into isolated workspaces or "spaces." These workspaces facilitate document indexing and querying through vector search and large language models (LLMs), while maintaining complete data privacy without external API calls. This development holds significance for the AI/ML community as it offers a hands-on tool for learning and experimentation with RAG systems in a secure environment. Notably, the system supports various document types, including PDFs and Microsoft Office files, as well as over 60 programming languages. Users can interact with real-time streaming responses and enjoy features like interactive visualizations and multi-session chat history. While it is not production-ready and should be used with caution, Local RAG Workspaces presents a promising step toward more private and effective AI applications in managing personal and professional projects.